3d spatially-resolved geometrical and functional models of ...66 recent advances in tissue staining,...

3D spatially-resolved geometrical and functional models of human liver tissue 1 reveal new aspects of NAFLD progression 2

3 Fabián Segovia-Miranda1, Hernán Morales-Navarrete1, Michael Kücken2, Vincent Moser3, 4 Sarah Seifert1, Urska Repnik1, Fabian Rost2,4, Alexander Hendriks5, Sebastian Hinz5, 5 Christoph Röcken5, Dieter Lüthjohann6, Yannis Kalaidzidis1, Clemens Schafmayer5, Lutz 6 Brusch2, Jochen Hampe3 and Marino Zerial1*. 7 8 9 1Max Planck Institute of Molecular Cell Biology and Genetics, Dresden, Germany. 10 2Center for Information Services and High Performance Computing, Technische Universität 11 Dresden, Dresden, Germany. 12 3Medical Department 1, University Hospital of the Technical University Dresden, Dresden, 13 Germany. 14 4Max Planck Institute for the Physics of Complex Systems, Dresden, Germany. 15 5University Hospital Schleswig-Holstein, Kiel, Germany. 16 6Institute of Clinical Chemistry and Clinical Pharmacology, University of Bonn, Bonn, 17 Germany. 18 19 20 21 22 *Correspondence: [email protected]; [email protected] 23

.CC-BY-NC-ND 4.0 International licensecertified by peer review) is the author/funder. It is made available under aThe copyright holder for this preprint (which was notthis version posted March 9, 2019. . https://doi.org/10.1101/572073doi: bioRxiv preprint

https://doi.org/10.1101/572073

http://creativecommons.org/licenses/by-nc-nd/4.0/

Abstract 24 25 Early disease diagnosis is key for the effective treatment of diseases. It relies on the 26 identification of biomarkers and morphological inspection of organs and tissues. 27 Histopathological analysis of human biopsies is the gold standard to diagnose tissue 28 alterations. However, this approach has low resolution and overlooks 3D structural changes 29 that are consequence of functional alterations. Here, we applied multiphoton imaging, 3D 30 digital reconstructions and computational simulations to generate spatially-resolved 31 geometrical and functional models of human liver tissue at different stages of non-alcoholic 32 fatty liver disease (NAFLD). We identified a set of new morphometric cellular parameters 33 correlated with disease progression. Moreover, we found profound topological defects in the 34 3D bile canaliculi (BC) network. Personalized biliary fluid dynamic simulations predicted an 35 increased pericentral biliary pressure and zonated cholestasis, consistent with elevated 36 cholestatic biomarkers in patients’ sera. Our spatially-resolved models of human liver tissue 37 can contribute to high-definition medicine by identifying quantitative multi-parametric cellular 38 and tissue signatures to define disease progression and provide new insights into NAFLD 39 pathophysiology. 40


https://doi.org/10.1101/572073


Introduction 41 42

High definition medicine is emerging as an integrated approach to profile and restore 43 the health of an individual using a pipeline of multi-parametric analytical and therapeutic 44 technologies1. High-definition medicine relies on large data sets, e.g. genomics, 45 metabolomics, to characterize human health at the molecular level. It also relies on imaging, 46 image analysis and computational modelling approaches to identify structural and functional 47 abnormalities in organs and tissues associated with a disease state. Histology has been 48 classically used to characterize tissue structure and remains the method of choice to 49 describe and monitor a large variety of pathologies2. However, this technique has several 50 disadvantages, e.g. it is subjective (depends on the pathologist’s skills), is often semi-51 quantitative and provides only two-dimensional (2D) information, i.e. does not account for 52 the three-dimensional (3D) complexity of tissues3. In recent years, an increasing number of 53 studies have highlighted the importance of considering 3D information for the 54 histopathological examination of tissues4-7. This is particularly crucial for the analysis of 3D 55 structures. The liver is a pertinent example of an organ with a complex 3D tissue 56 organization8. It consists of functional units, the liver lobule9,10, containing two intertwined 57

networks, the sinusoids for blood flow and the bile canaliculi (BC) for bile secretion and flux9. 58 Sinusoids and BC run antiparallel along the central vein (CV)-portal vein (PV) axis. The 59 hepatocytes are the major parenchymal cells and display a peculiar and unique type of cell 60 polarity distinct from that of simple epithelia11. Whereas in epithelia all cells share the same 61 orientation with their apical surface facing the lumen of the organ, hepatocytes are 62 sandwiched between the sinusoidal endothelial cells and share the apical surface with 63 multiple neighbouring hepatocytes to form a 3D BC network12,13. Such an architecture makes 64 it difficult to grasp the 3D organization of cells and tissue from 2D histological sections. 65 Recent advances in tissue staining, optical clearing and multi-photon microscopy allow 66 imaging thick sections of tissues such that 3D information can be captured14,15. Computer 67 software process the images to generate 3D digital reconstructions of tissues, i.e. 68 geometrical models8, with single-cell resolution. These provide a detailed quantitative 69 description of the different cells and structures forming the tissue. The geometrical 70 information extracted from the 3D reconstruction can also be used to generate predictive 71 models of tissue function e.g. biliary fluid dynamic16, thus gaining novel insights into liver 72 tissue organization and function. Thereby, geometrical models can be used to quantitatively 73


https://doi.org/10.1101/572073


describe the tissue architecture and function, improving our understanding of liver biology 74 and pathobiology. 75

Non-alcoholic fatty liver disease (NAFLD), defined as an accumulation of 76 triglycerides and lipid droplets (LD) in the liver in absence of alcohol intake (infiltration in > 77 5% of hepatocytes), is rising to the most common chronic liver disease worldwide17. NAFLD 78 includes a spectrum of liver diseases, ranging from simple steatosis to non-alcoholic 79 steatohepatitis (NASH)17. Whereas steatosis is considered as a “non-progressive” status of 80 the disease, NASH has the potential to progress to more severe stages, such as cirrhosis 81 or hepatocellular carcinoma, leading eventually to liver failure and transplantation3,17. Thus, 82 the understanding of the transition from steatosis to NASH as a disease-defining moment 83 for NAFLD prognosis is key to a deeper understanding of disease pathophysiology. Liver 84 biopsy and histological inspection of thin tissue slices (< 10 µm) constitute the current gold 85 standard for the diagnosis of steatosis and NASH2,17,18. Unfortunately, due to the limitations 86

in providing 3D information, alterations in 3D tissue structures such as BC3,17, have been 87 overlooked. Therefore, new approaches are required to gain a more complete 88 understanding of NAFLD establishment and its progression to NASH. In this study, we 89 generated 3D spatially resolved geometrical and functional models of human liver tissue for 90 different stages of NAFLD to contribute to a high definition medical diagnosis of disease 91 establishment and progression. 92


https://doi.org/10.1101/572073


Results 93 94 Tri-dimensional geometrical models of human liver tissue 95

To identify parameters that can quantitatively discriminate the transition from simple 96 steatosis to early NASH (eNASH) we stained, imaged and digitally reconstructed human 97 liver tissue in 3D. We focused on cell and nuclear morphology, LD, and tissue features such 98 as BC and sinusoidal networks, and their spatial distribution along the CV-PV axis. For this 99 we tested 26 antibodies combinatorially with dyes and antigen retrieval protocols 100 (Supplementary Table 2). Our final pipeline (see Methods for details) includes the following 101 steps. First, 100 µm liver sections were heated and antigen retrieved using citric acid buffer 102 (CAAR). Second, we stained for BC (CD13), sinusoids (fibronectin), nuclei (DAPI), lipid 103 droplets LD (BODIPY) and cell borders (LDLR), optically cleared with SeeDB19 and imaged 104 at high resolution using multiphoton microscopy (0.3 µm x 0.3 µm x 0.3 µm per voxel) 105 (Extended Data Fig. 1). Because this protocol did not provide a good cell border staining of 106 the pericentral hepatocytes in STEA and eNASH, for cell-based measurements (see below 107 Fig. 3), we used an alternative protocol without antigen retrieval enabling the staining of 108 sinusoids (fibronectin), nuclei (DAPI), LD (BODIPY) and cell borders (phalloidin). We applied 109 this pipeline to biopsies from twenty-two patients classified into four groups: normal control 110 (NC, n = 6), healthy obese (HO, n = 4), steatosis (STEA, n = 5) and early NASH (eNASH, n 111 = 7). The demographic, clinical and histological details of the samples are summarized in 112 the Supplementary Table 1. All images cover one complete CV-PV axis within a liver lobule. 113 Finally, we reconstructed the various stained components of the tissue using our open-114 source software Motion Tracking (http://motiontracking.mpi-cbg.de) as described8 (Fig. 1 115 and Supplementary Video 1-2). The generation of geometrical models constituted the basis 116 for the quantitative and structural characterization of the different components forming the 117 liver tissue in the NAFLD biopsies. 118

Nuclear-based analysis of NAFLD progression 119

We first quantified properties of hepatocytes nuclei, such as cell nuclearity and 120 ploidy, since hepatocytes are heterogenous in both mouse and human8,20. We quantified 121 nuclear vacuolization/glycogenation given it is a common histological characteristic in 122 NAFLD linked to insulin resistance21,22. Finally, we measured nuclear texture homogeneity, 123 a feature associated with various pathological conditions, such as cancer, inflammation, 124


https://doi.org/10.1101/572073


cardiomyopathy, etc23-26, with methylation and acetylation status27 and, more recently, with 125

transcriptional activity28. We neither observed differences in the proportion of 126 mono/binuclear cells nor in the ploidy between the groups (Extended Data Fig. 2a and b). 127 The average values of several nuclear features showed only modest variations Extended 128 Data Fig. 2). However, many functional and morphological features of the liver change along 129 the CV-PV axis, such as metabolic zonation29,30, ploidy8, cell volume8, BC16. Therefore, to 130 account for potential morphological heterogeneities along the CV-PV axis, we 131 computationally divided this axis into ten equidistant zones (Extended Data Fig. 2c) 132 (Methods). We found major differences in nuclear elongation around the CV and 133 vacuolization around the PV between the different groups (Extended Data Fig. 2d,e). 134 Moreover, we identified zonated and progressive differences in nuclear homogeneity as 135 disease progresses (Extended Data Fig. 2f-i). Therefore, our analysis reveals that, in spite 136 of modest changes in the average values, the combined zonated values of nuclear 137 vacuolization and texture homogeneity allow discriminating between all four patient groups. 138

Morphometric parameters of LD correlate with disease progression 139

The finding that quantitative spatially-resolved analysis of nuclear parameters can 140 reveal changes that are not evident in an average estimate prompted us to re-evaluate the 141 morphometric characterization of LD. Even though LD are the most typical hallmark of 142 NAFLD, a detailed quantitative description of their size and their spatial localization within 143 the liver lobule has not been achieved yet. Contrary to traditional histology3, immunostaining 144 of thick tissue sections preserved most of the LD (Fig 2a). In agreement with the 145 histopathological description of NAFLD3,17, a major increase in LD was observed in STEA 146 and eNASH, which were concentrated between the second and the fifth zones (Fig. 2b). 147 However, the LD occupy a higher volume of the tissue in eNASH than STEA. It is known 148 that the LD can present massive differences in size31,32. The LD in the human liver samples 149

ranged from ~1 µm3 to 20,000 µm3 (Fig. 2c). To inquire whether differences in LD size may 150

correlate with the disease state, we performed a population analysis based on their volume 151 distributions (Fig. 2c). We defined three sub-populations of LD, namely, small (< 14 µm3), 152 medium (14 – 400 µm3) and large (> 400 µm3) ones (Fig. 2d-f). Strikingly, we found that the 153 three sub-populations varied between disease conditions. Whereas small LD were evenly 154 distributed along the CV-PV in all conditions, large LD were highly enriched towards the 155 pericentral zone in STEA and eNASH, occupying up to 25% of the tissue volume (Fig. 2f). 156


https://doi.org/10.1101/572073


Most importantly however, medium LD were mostly present in eNASH and spanning the 157 entire liver lobule (Fig. 2e), hence contributing to discriminate between STEA and eNASH 158 in the periportal area (Fig. 2e). 159

The zonated increase in LD size during disease progression is such that some LD in 160 the pericentral zone become even larger than a normal hepatocyte. This leads to global 161 changes in tissue structure. To quantify such changes, we measured the spatial distribution 162 of cell density, cell volume and percentage of cell volume occupied by LD. We found ∼50% 163 reduction in the number of hepatocytes located between the CV and the middle zone in 164 STEA and eNASH, when compared with NC and HO samples (Fig. 3a). This reduction was 165 compensated by a massive increase in cell volume (Fig. 3b). Hepatocytes were two times 166 larger than the average size (Fig. 3b), reaching values up to up to ~100,000 µm3 for STEA 167 and NASH (ten times bigger than a small hepatocyte) (Fig. 3c). A population analysis of the 168 hepatocytes based on their volume revealed a characteristic distribution of different cells 169 populations along the liver lobule (Fig. 3c-f). STEA and eNASH were characterized 170 predominantly by small and large hepatocytes which are anti-correlated along CV-PV axis 171 (Fig. 3d-f). Even though cell density and cell volume were practically indistinguishable 172 between STEA and eNASH, we observed a remarkable phenotype regarding the fraction of 173 cell volume occupied by LD (Fig. 3g-i). In eNASH, hepatocytes accumulated LD even in the 174 periportal zone (Fig. 3h, 3i and Supplementary Video 3), suggesting that LD accumulation 175 progressively extends to the PV as the disease progresses. 176

Altogether, these data reveal profound quantitative morphological disparities in cell 177 size and LD content along the CV-PV axis between NC, HO, STEA and eNASH. Specifically, 178 the percentage of cell volume occupied by LD in the PV and CV zones can serve to 179 discriminate STEA and eNASH. 180

Alterations in apical protein trafficking 181

The massive presence of LD that occupy a large portion of the cytoplasm raises the 182 question of whether trafficking of proteins to the apical plasma membrane of hepatocytes is 183 affected. We analysed the localization of four apical proteins, aminopeptidase N (CD13), 184 bile salt export pump (BSEP), multidrug resistant-associated protein (MRP2) and 185 dipeptidylpeptidase 4 (DPPIV). CD13, BSEP and MRP2 were correctly localized to the 186 apical membrane in all conditions (Extended Data Fig. 1, 3a-b). DPPIV was enriched on the 187 apical membrane with a small fraction on the basal membrane in NC and HO (Extended 188


https://doi.org/10.1101/572073


Data Fig. 3c). Strikingly, DPPIV was redistributed to the lateral membrane in pericentral 189 hepatocytes in STEA and eNASH, whereas it retained its normal localization on the 190 periportal zone (Extended Data Fig. 3c). Considering that DPPIV follows the transcytotic 191 route to the apical surface33,34 whereas BSEP and MRP2 do not34-36, the mislocalization of 192 DPPIV suggests a possible disruption of some stage of endocytosis and transcytosis of this 193 cargo molecule in the pericentral hepatocytes. This supports previous findings regarding the 194 misregulation of membrane protein trafficking in NAFLD37 and prompted us to evaluate 195 whether the integrity of the BC could be affected during the disease progression. 196

Bile canaliculi network shows geometrical and topological-zonated defects in NAFLD 197

To determine whether 3D structures such as the BC and sinusoidal networks are 198 affected, we carried out a geometrical and topological characterization of both networks. 199 Even though we observed a slight reduction in the total length of the sinusoidal network in 200 STEA and eNASH (Extended Data Fig. 4e), no major defects in sinusoidal microanatomy 201 were detected (Extended Data Fig. 4a-d, f-g) (volume fraction, radius, branching and 202 connectivity). Next, we analysed the BC network. Contrary to the very packed and 203 homogeneous appearance in NC and HO, the BC in STEA and eNASH displayed clear 204 morphological defects which were more pronounced in the pericentral zone (Fig. 4a, b). A 205 more detailed analysis revealed a sustained increase in BC radius in NASH throughout the 206 CV-PV axis (Fig. 4d). In addition, in both STEA and eNASH, we observed a strong reduction 207 in the total length of the BC towards the pericentral zone (Fig. 4a and f). Other geometrical 208 properties of the BC network, such as volume fraction and junction density were unaffected 209 (Fig. 4c and e). 210

Finally, to investigate the topological properties of the BC network, we performed an 211 analysis of network connectivity (see Methods). Surprisingly, we found a pronounced decay 212 in the connectivity in STEA and eNASH towards the pericentral region (Fig. 4a, b, g and h, 213 Supplementary Video 4). One possibility is that the alterations in BC may be the 214 consequence of the spatial constrains arising from the presence of large cells in the tissue 215 (Fig. 3f). To test this, we inspected the connectivity of the sinusoidal network. We found that 216 no defect was observed for the sinusoidal network (Extended Data Fig. 4a, f and g), 217 supporting the idea that the BC network is specifically affected and not an indirect 218 consequence of spatial constraints. Thus, our data point at specific geometrical and 219 topological alterations in the pericentral BC network in both STEA and eNASH. 220


https://doi.org/10.1101/572073


Personalised model of bile flow predicts increase in bile pressure in the pericentral 221 zone 222

The observed alterations of BC network architecture are likely to have consequences 223 for liver tissue function, particularly for bile flow. Clearly, information on bile velocity and 224 pressure in disease conditions could be insightful. However, it is not yet possible to measure 225 bile flow in the human liver at the level of BC as in animal models. We recently developed a 226 computational model of bile fluid dynamics, validated its quantitative predictions in mouse 227 models and demonstrated that bile velocity and bile pressure distributions along the liver 228 lobule strongly depend on BC geometry16. However, this model16 is not suitable to handle 229 the extreme inhomogeneity of BC density such as the ones encountered in tissue distorted 230 by the presence of large LD (i.e. in STEA and eNASH). Therefore, we addressed this issue 231 and further developed our model in a spatially-resolved fashion (Fig. 5a). Shortly, the refined 232 model is based on conservation of mass for water and osmolytes and Darcy’s Law for 233 laminar flow. The proportionality constant in Darcy’s Law was derived from the porous media 234 theory. Boundary conditions were set to zero velocity at the outer surface of the central vein, 235 and ambient pressure at the portal outlet. Since we obtained morphometric data for liver 236 tissue from individual patients, we aimed at developing personalised models, i.e. 237 parameterized by individual geometrical measurements (BC volume fraction eBC, BC radius 238

rBC., fraction of connected BC, canaliculi tortuosity t, apical surface density A, intra-239

canalicular volume fraction occupied by microvilli 1-a) (Fig. 4c, 4d, 4g, and Extended Data 240 Fig. 5) and previously reported values (viscosity, permeability and osmolyte secretion rate). 241 No free parameters remained and, hence, no parameter fitting was needed (see Methods 242 for details). Next, we applied this model to predict bile velocity, pressure and solute 243 concentration distributions across the liver lobule for individual patients liver tissue 3D 244 reconstructions from all the four groups. 245

The model predicts bile velocities in the periportal area of about 1.2 ± 0.4 μm/sec for 246 all patient groups (Fig. 5b-e). Very similar velocities have been reported in mouse16. 247 However, the predicted pressure in the pericentral area differed significantly between the 248 patient groups. In the NC and HO groups this pressure was predicted to be lower than 1500 249 Pa in all patients (963.2 ± 285.2 Pa, mean ± SD) (Fig. 5b-c). This is consistent with the 250 reported maximum biliary secretion pressure of 1,000–1,500 Pa in the extrahepatic biliary 251 system in rats38. In the STEA and eNASH groups, the model predicted an abrupt increase 252


https://doi.org/10.1101/572073


of bile pressure toward the pericentral zone (Fig. 5d-e). For six STEA and eNASH patients 253 (55%), the pericentral pressure exceeded 1,500 Pa and for four patients (36%) it exceeded 254 twice the maximum pressure predicted for NC (3000 Pa) (Fig. 5d-e). The compensatory 255 effect in bile pressure observed in eNASH is mostly due to the dilation of BC (Fig. 4d). 256 Therefore, our model predicts an increase in pericentral bile pressure in STEA and eNASH 257 conditions ranging from relatively mild to quite severe, depending on the BC geometry of 258 individual patients. 259

We next set out to validate the model predictions. As it is impossible to measure bile 260 flow and pressure in the human liver, we considered possible consequences of changes in 261 bile pressure. Increased bile pressure is a hallmark of cholestasis39,40. Therefore, as readout 262 of increased bile pressure, we analysed the most commonly used cholestatic biomarkers in 263 serum, including bilirubin, gamma glutamyl transpeptidase (GGT), alkaline phosphatase 264 (ALP) and BAs. To increase the statistical power, we analysed additional sera samples for 265 the different groups (NC = 25, HO = 25, STEA = 24 and eNASH = 26 samples). Whereas 266 we found that both bilirubin and GGT were elevated in STEA and eNASH (Extended Data 267 Fig. 6a-b), we did not detect significant changes in the levels of ALP, total BAs and primary 268 BAs between the groups (Extended Data Fig. 6c, 7b,d). Strikingly, when we analysed the 269 correlation between the predicted pericentral bile pressure and the biomarkers for individual 270 patients from all groups, we found a strong correlation for the majority of the cholestatic 271 biomarkers, with GGT having the strongest correlation (Pearson correlation coefficient ALP= 272 0.473, total BAs 0.505, primary BA 0.518 and GGT 0.680; (Extended Data Fig. 8a-e and 273 Fig. 5f). In contrast, aspartate aminotransferase (AST) and alanine aminotransferase (ALT), 274 biomarkers of hepatocellular liver damage which are not increased in cholestasis, showed 275 no correlation (Extended Data Fig. 8f-g). The presence of pericentral cholestasis was also 276 supported by the increase in the predicted pericentral concentration of BAs (proportional to 277 the lumped concentration of all osmolytes. Extended Data Fig. 9), which in combination with 278 high pressure and low flow velocity, are signs of a defective bile flow. Altogether, our model 279 predicts a significant degree of zonated cholestasis as a new component of the NAFLD 280 pathophysiology. 281


https://doi.org/10.1101/572073


Discussion 282 283

High definition medicine provides a novel approach to understand human health of 284

individuals with unprecedented precision1. One of its pillars is the combination of image 285

analysis and computational modelling to uncover tissue alterations at different structural and 286 functional levels during disease progression. During the last years, there has been an 287 enormous interest in getting a better understanding of NAFLD establishment and 288 progression due to its growing impact on public health41. A lot of attention has been mostly 289 drawn to the role of signalling pathways42-44, microbiome45,46, metabolism47, genetic risk 290

factors48, BAs44, etc. However, a major challenge is to understand how the molecular 291 alterations detected are expression of the organ dysfunction, manifested as morphological 292 and functional alterations of cells and tissue architecture. 293

The classical histological analysis has provided insights into fundamental aspects of 294 NAFLD. However, a quantitative description of the 3D tissue morphology is indispensable, 295 particularly for the liver which contains intertwined 3D networks enabling the flow of fluids, 296 the sinusoids for blood flow and the BC for bile secretion and flux9. Here, we used high 297 resolution multiphoton microscopy and 3D digital reconstructions to generate a comparative 298 dataset of structural changes of human liver tissue from NC, HO, STEA and eNASH patients. 299 We identified a set of zonated morphological alterations that correlate with disease 300 progression, such as a characteristic size distribution of LD and nuclear texture 301 homogeneity, that can be used as tissue biomarkers to distinguish between different stages 302 of NAFLD progression. In addition, the 3D digital reconstruction provided the first evidence 303 that BC integrity is disrupted during NAFLD progression, bringing BC integrity and the 304 mechanisms involved in its maintenance and homeostasis (cell polarity, trafficking, bile flow, 305 BAs turnover, etc.) into focus for NAFLD studies. Based on the geometrical and topological 306 information extracted from the BC, we used a computational personalised model to connect 307 the microanatomy of BC with biliary fluid dynamics within the lobule. Our model predicted 308 high bile pressure in the pericentral area and a significant degree of zonated cholestasis in 309 STEA and eNASH patients, a prediction that was validated by the detection of cholestatic 310 biomarkers in serum. Our data show that geometrical models of human tissues coupled to 311 computational modelling is a powerful strategy to describe human physiology and 312 physiopathology. 313

The spatially-resolved quantitative analysis of the 3D reconstructions of human liver 314


https://doi.org/10.1101/572073


samples revealed a set of unknown morphological features, ranging from the (sub)cellular 315 (nuclear texture, LD content, polarity) to the tissue level (BC integrity), that are perturbed 316 during NAFLD progression. First, we detected changes in nuclear texture, which have been 317 reported in several diseases24,25,49, but have not yet been studied in NAFLD. Indeed, 318 changes in nuclear texture homogeneity in the pericentral hepatocytes may reflect changes 319 in transcriptional activity27,28 and could serve as a new component of the histological scores 320 of NAFLD progression. Second, although the accumulation of LD is a characteristic feature 321 of NAFLD, our analysis revealed quantitative changes in their size distribution, with the 322 medium LD mostly present in eNASH and spanning the entire liver lobule. In healthy 323 conditions, the LD number and size are accurately regulated50 and changes in LD 324 distribution point at specific alterations in the mechanisms regulating LD biogenesis and 325 catabolism. Third, and most striking, we observed alterations of the apical plasma 326 membrane of hepatocytes and of the BC network. The pericentral hepatocytes showed 327 mislocalization of DPPIV, pointing towards a dysregulation in apical protein trafficking37. 328 Interestingly, not all apical proteins were missorted, suggesting that trafficking defects could 329 be pathway- (transcytosis) and/or cargo-specific. Such defects in protein trafficking 330 correlated with the reduction of BC connectivity in the pericentral zone. This is the first time 331 that geometrical and topological properties of the BC are studied in human liver tissue from 332 NAFLD biopsies. The unaltered architecture of the sinusoidal network along the different 333 stages of NAFLD rules out the possibility that this reduction in BC connectivity is simply due 334 to the spatial constrains imposed by the presence of the large pericentral hepatocytes. 335 These results pose the question of how the alterations in apical surface of the hepatocytes 336 collectively result in the compromised connectivity of the BC network. 337

The altered BC microanatomy and the consequent increase in bile pressure in the 338 pericentral zone suggest that STEA and eNASH livers are affected by a pericentral 339 cholestasis which may contribute to the changes in BA composition observed in 340 NAFLD42,44,51. Unimpaired bile flow is essential for normal liver function. Previous studies 341 have documented that bile accumulation, due to its detergent-like properties, can cause liver 342 damage52,53 and bile pressure can affect metabolism54. The occurrence of zonated 343 cholestasis is a new piece in the NAFLD physiopathology puzzle that contributes to clarify 344 some aspects of the disease so far without explanation, e.g. increase of GGT levels55, bile 345 acids in serum56,57, upregulation of MRP3 in NASH58,59 and the beneficial effect of UDCA 346

treatment in NAFLD60, all sign of an ongoing cholestasis40,61,62. 347


https://doi.org/10.1101/572073


In recent years, a lot of research has been devoted to the role of BAs and the activity 348 of their receptor FXR, in NAFLD42-44. However, there is currently no explanation for the 349 alterations in BAs composition in blood, the decreased ratio of secondary/primary BAs 350 observed by us (Extended Data Fig. 7) and others42,44,51, and whether it correlates with 351 changes in tissue morphology44. Our data shed new light on this problem. Our results 352 suggest that the altered BC microanatomy leading to increased bile pressure in the 353 pericentral zone may hamper the ongoing bile acid secretion into BC, as apical pumps 354 (BSEP, MRP2) have to operate against elevated luminal BA concentrations (Extended Data 355 Fig. 9). This could lead to back-flux of primary BA into the blood (Extended Data Fig. 8d), 356 reducing the availability of primary bile acids to be converted into secondary bile acids by 357 the microbiota in the intestine (Extended Data Fig. 7e). 358

The combination of experimental data with computational models of tissues has 359 proven successful in elucidating pathogenetic mechanisms using animal models16,63,64. 360

However, animal models very often fail to mimic human diseases65, including NAFLD66. In 361 this study, the geometrical models of liver tissue from human biopsies combined with 362 spatially-resolved computational simulations revealed new aspects of NAFLD pathology. 363 This approach may help to identify biomarkers for early disease diagnosis and predict the 364 functional status of the tissue with potential applications in high-definition medicine1,67,68. 365

366


https://doi.org/10.1101/572073


Acknowledgements 367

We are grateful to Oleksandr Ostrenko, Juan Francisco Miquel Poblete and Sophie Nehring 368 for fruitful discussions, and Sebastian Bundschuh for helping to set up the 2-photon 369 microscope. We thank the Center for Information Services and High Performance 370 Computing (ZIH) of the TU Dresden for the generous provision of computing power. We 371 would also like to thank the following Services and Facilities of the Max Planck Institute of 372 Molecular Cell Biology and Genetics for their support: Light Microscopy Facility (LMF) and 373 the Electron Microscopy Facility. 374

This work was financially supported by the German Federal Ministry of Education and 375 Research (BMBF) (LiSyM: grant #031L0038 to M.Z., grant #031L0033 to L.B., grant 376 #031L0031 to J.H., DYNAFLOW: grant #031L0082B to M.Z., grant #031L008A to L.B. and 377 SYSBIO II: grant #031L0044 to M.Z.), European Research Council (ERC) (grant #695646 378 to M.Z.) and the Max Planck Society (MPG). 379

Author contributions 380

F.S-M., J.H. and M.Z. conceived the project. F.S-M., V.M. and S.S. performed the 381 immunofluorescence experiments and imaging. H.M-N. and Y.K. developed the image 382 analysis algorithms. F.S-M., V.M. and H.M-N performed the 3D tissue reconstructions. H.M-383 N. and F.S-M. performed the data analysis and interpretation of the results. U.R. performed 384 the electron microscopy. A.H., S.H., C.R., and C.S obtained the samples and characterized 385 the patients. D.L. measured bile acids. M.K., F.R., Y.K. and L.B. conceived and developed 386 the mathematical model. M.K and F.R. programmed and simulated the mathematical model 387 and performed statistical analysis. M.K. and L.B. interpreted results and wrote the model 388 description. F.S-M., H.M-N., M.K., Y.K., L.B. and M.Z. wrote the manuscript. 389

Competing interests 390 391 Authors declare no competing interests 392


https://doi.org/10.1101/572073


References 393

1. Torkamani, A., Andersen, K. G., Steinhubl, S. R. & MD, E. J. T. High-Definition 394 Medicine. Cell 170, 828–843 (2017). 395

2. Mills, S. E. Histology for Pathologists. (Lippincott Williams & Wilkins, 2012). 396 3. Nalbantoglu, I. L. K. & Brunt, E. M. Role of liver biopsy in nonalcoholic fatty liver 397

disease. World J. Gastroenterol. 20, 9026–9037 (2014). 398 4. Tanaka, N. et al. Whole-tissue biopsy phenotyping of three-dimensional tumours 399

reveals patterns of cancer heterogeneity. Nature Biomedical Engineering 1–14 400 (2017). doi:10.1038/s41551-017-0139-0 401

5. Hägerling, R. et al. VIPAR, a quantitative approach to 3D histopathology applied to 402 lymphatic malformations. JCI Insight 2, 1–14 (2017). 403

6. Nojima, S. et al. CUBIC pathology: three- dimensional imaging for pathological 404 diagnosis. Sci Rep 1–14 (2017). doi:10.1038/s41598-017-09117-0 405

7. Glaser, A. K. et al. Light-sheet microscopy for slide-free non-destructive pathology 406 of large clinical specimens. Nature Biomedical Engineering 1, 0084 (2017). 407

8. Morales-Navarrete, H. et al. A versatile pipeline for the multi-scale digital 408 reconstruction and quantitative analysis of 3D tissue architecture. eLife 4, 841 409 (2015). 410

9. Elias, H. A re-examination of the structure of the mammalian liver; the hepatic lobule 411 and its relation to the vascular and biliary systems. Am. J. Anat. 85, 379–456– 15 pl 412 (1949). 413

10. Elias, H. LIVER MORPHOLOGY. Biological Reviews 30, 263–310 (1955). 414 11. Morales-Navarrete, H. et al. Liquid-crystal organization of liver tissue. bioRxiv 415

495952 (2018). doi:10.1101/495952 416 12. Elias, H. A re-examination of the structure of the mammalian liver; parenchymal 417

architecture. Am. J. Anat. 84, 311–333 (1949). 418 13. Treyer, A. & Müsch, A. Hepatocyte Polarity. (John Wiley & Sons, Inc., 2013). 419

doi:10.1002/cphy.c120009 420 14. Tainaka, K., Kuno, A., Kubota, S. I., Murakami, T. & Ueda, H. R. Chemical 421

Principles in Tissue Clearing and Staining Protocols for Whole-Body Cell Profiling. 422 Annu Rev Cell Dev Biol 32, 713–741 (2016). 423

15. Richardson, D. S. & Lichtman, J. W. Clarifying Tissue Clearing. Cell 162, 246–257 424 (2015). 425

16. Meyer, K. et al. A Predictive 3D Multi-Scale Model of Biliary Fluid Dynamics in the 426 Liver Lobule. Cell Systems 4, 277–290.e9 (2017). 427

17. Hardy, T., Oakley, F., Anstee, Q. M. & Day, C. P. Nonalcoholic Fatty Liver Disease: 428 Pathogenesis and Disease Spectrum. Annu Rev Pathol 11, 451–496 (2016). 429

18. Bedossa, P. Pathology of non-alcoholic fatty liver disease. Liver Int 37, 85–89 430 (2017). 431

19. Ke, M.-T., Fujimoto, S. & Imai, T. SeeDB: a simple and morphology-preserving 432 optical clearing agent for neuronal circuit reconstruction. Nature Neuroscience 16, 433 1154–1161 (2013). 434

20. Wang, M.-J., Chen, F., Lau, J. T. Y. & Hu, Y.-P. Hepatocyte polyploidization and its 435 association with pathophysiological processes. 8, e2805–7 (2017). 436

21. Younossi, Z. M. et al. Nonalcoholic fatty liver disease: assessment of variability in 437 pathologic interpretations. Mod. Pathol. 11, 560–565 (1998). 438


https://doi.org/10.1101/572073


22. Levene, A. P. & Goldin, R. D. Physiological hepatic nuclear vacuolation--how long 439 does it persist? Histopathology 56, 426–429 (2010). 440

23. Hamilton, P. W. et al. Digital pathology and image analysis in tissue biomarker 441 research. Methods 70, 59–73 (2014). 442

24. Bartels, P. H. et al. Chromatin texture signatures in nuclei from prostate lesions. 443 Anal. Quant. Cytol. Histol. 20, 407–416 (1998). 444

25. Jørgensen, T., Yogesan, K., Tveter, K. J., Skjørten, F. & Danielsen, H. E. Nuclear 445 texture analysis: a new prognostic tool in metastatic prostate cancer. Cytometry 24, 446 277–283 (1996). 447

26. Doudkine, A., Macaulay, C., Poulin, N. & Palcic, B. Nuclear texture measurements 448 in image cytometry. Pathologica 87, 286–299 (1995). 449

27. Orr, J. A. & Hamilton, P. W. Histone acetylation and chromatin pattern in cancer. A 450 review. Anal. Quant. Cytol. Histol. 29, 17–31 (2007). 451

28. Ricci, M. A., Cosma, M. P. & Lakadamyali, M. Super resolution imaging of 452 chromatin in pluripotency, differentiation, and reprogramming. Curr. Opin. Genet. 453 Dev. 46, 186–193 (2017). 454

29. Felmlee, D. J., Grün, D. & Baumert, T. F. Zooming in on liver zonation. Hepatology 455 67, 784–787 (2018). 456

30. Gebhardt, R. & Matz-Soja, M. Liver zonation: Novel aspects of its regulation and its 457 impact on homeostasis. World J. Gastroenterol. 20, 8491–8504 (2014). 458

31. Kleiner, D. E. & Makhlouf, H. R. Histology of Nonalcoholic Fatty Liver Disease and 459 Nonalcoholic Steatohepatitis in Adults and Children. Clinics in Liver Disease 20, 460 293–312 (2016). 461

32. Kochan, K. et al. Raman spectroscopy analysis of lipid droplets content, distribution 462 and saturation level in Non-Alcoholic Fatty Liver Disease in mice. J. Biophotonics 8, 463 597–609 (2015). 464

33. Wang, L. & Boyer, J. L. The maintenance and generation of membrane polarity in 465 hepatocytes. Hepatology 39, 892–899 (2004). 466

34. Zeigerer, A. et al. Rab5 is necessary for the biogenesis of the endolysosomal 467 system in vivo. Nature 485, 465–470 (2012). 468

35. Wakabayashi, Y., Dutt, P., Lippincott-Schwartz, J. & Arias, I. M. Rab11a and myosin 469 Vb are required for bile canalicular formation in WIF-B9 cells. Proc Natl Acad Sci 470 USA 102, 15087–15092 (2005). 471

36. Kipp, H., Pichetshote, N. & Arias, I. M. Transporters on demand: intrahepatic pools 472 of canalicular ATP binding cassette transporters in rat liver. J Biol Chem 276, 7218–473 7224 (2001). 474

37. Dzierlenga, A. L. & Cherrington, N. J. Misregulation of membrane trafficking 475 processes in human nonalcoholic steatohepatitis. J Biochem Mol Toxicol 37, 476 e22035–7 (2018). 477

38. Weis, E. E. & Barth, C. A. The extracorporeal bile duct: a new model for 478 determination of bile flow and bile composition in the intact rat. J. Lipid Res. 19, 479 856–862 (1978). 480

39. Jansen, P. L. M. et al. The ascending pathophysiology of cholestatic liver disease. 481 Hepatology 65, 722–738 (2017). 482

40. Pollock, G. & Minuk, G. Y. Diagnostic considerations for cholestatic liver disease. J 483 Gastroenterol Hepatol 32, 1303–1309 (2017). 484

41. Friedman, S. L., Neuschwander-Tetri, B. A., Rinella, M. & Sanyal, A. J. Mechanisms 485 of NAFLD development and therapeutic strategies. Nat Med 24, 908–922 (2018). 486


https://doi.org/10.1101/572073


42. Chow, M. D., Lee, Y.-H. & Guo, G. L. The role of bile acids in nonalcoholic fatty liver 487 disease and nonalcoholic steatohepatitis. Molecular Aspects of Medicine 1–11 488 (2017). doi:10.1016/j.mam.2017.04.004 489

43. Arab, J. P., Karpen, S. J., Dawson, P. A., Arrese, M. & Trauner, M. Bile acids and 490 nonalcoholic fatty liver disease: Molecular insights and therapeutic perspectives. 491 Hepatology 65, 350–362 (2016). 492

44. Jahn, D. & Geier, A. Bile acids in NASH: pathophysiological driving force or 493 innocent bystanders? Hepatology 1–8 (2017). doi:10.1002/hep.29543 494

45. Henao-Mejia, J. et al. Inflammasome-mediated dysbiosis regulates progression of 495 NAFLD and obesity. Nature 482, 179–185 (2012). 496

46. Betrapally, N. S., Gillevet, P. M. & Bajaj, J. S. Changes in the Intestinal Microbiome 497 and Alcoholic and Nonalcoholic Liver Diseases: Causes or Effects? 498 Gastroenterology 150, 1745–1755.e3 (2016). 499

47. Perla, F. M., Prelati, M., Lavorato, M., Visicchio, D. & Anania, C. The Role of Lipid 500 and Lipoprotein Metabolism in Non-Alcoholic Fatty Liver Disease. Children (Basel) 501 4, 46 (2017). 502

48. Romeo, S. et al. Genetic variation in PNPLA3 confers susceptibility to nonalcoholic 503 fatty liver disease. Nat Genet 40, 1461–1465 (2008). 504

49. Huisman, A. et al. Discrimination between benign and malignant prostate tissue 505 using chromatin texture analysis in 3-D by confocal laser scanning microscopy. 506 Prostate 67, 248–254 (2007). 507

50. Walther, T. C., Chung, J. & Farese, R. V., Jr. Lipid Droplet Biogenesis. Annu Rev 508 Cell Dev Biol 33, 491–510 (2017). 509

51. Puri, P. et al. The presence and severity of nonalcoholic steatohepatitis is 510 associated with specific changes in circulating bile acids. Hepatology 67, 534–548 511 (2017). 512

52. Trauner, M., Meier, P. J. & Boyer, J. L. Molecular pathogenesis of cholestasis. N 513 Engl J Med 339, 1217–1227 (1998). 514

53. Coleman, R., Iqbal, S., Godfrey, P. P. & Billington, D. Membranes and bile 515 formation. Composition of several mammalian biles and their membrane-damaging 516 properties. Biochem J 178, 201–208 (1979). 517

54. Jansen, P. L. M. Hydrodynamics of bile flow: Lessons from computational modeling. 518 Hepatology 67, 1624–1627 (2018). 519

55. Chan, T. T. & Wong, V. W. S. In search of new biomarkers for nonalcoholic fatty 520 liver disease. Clinical Liver Disease 8, 19–23 (2016). 521

56. Ferslew, B. C. et al. Altered Bile Acid Metabolome in Patients with Nonalcoholic 522 Steatohepatitis. Digestive Diseases and Sciences 60, 3318–3328 (2015). 523

57. Mouzaki, M. et al. Bile Acids and Dysbiosis in Non-Alcoholic Fatty Liver Disease. 524 PLoS ONE 11, e0151829–13 (2016). 525

58. Canet, M. J. et al. Altered Regulation of Hepatic Efflux Transporters Disrupts 526 Acetaminophen Disposition in Pediatric Nonalcoholic Steatohepatitis. Drug Metab 527 Dispos 43, 829–835 (2015). 528

59. Hardwick, R. N., Fisher, C. D., Canet, M. J., Scheffer, G. L. & Cherrington, N. J. 529 Variations in ATP-binding cassette transporter regulation during the progression of 530 human nonalcoholic fatty liver disease. Drug Metab Dispos 39, 2395–2402 (2011). 531

60. Steinacher, D., Claudel, T. & Trauner, M. Therapeutic Mechanisms of Bile Acids 532 and Nor-Ursodeoxycholic Acid in Non-Alcoholic Fatty Liver Disease. Dig Dis 35, 533 282–287 (2017). 534


https://doi.org/10.1101/572073


61. Chai, J. et al. Elevated hepatic multidrug resistance-associated protein 3/ATP-535 binding cassette subfamily C 3 expression in human obstructive cholestasis is 536 mediated through tumor necrosis factor alpha and c-Jun NH2-terminal 537 kinase/stress-activated protein kinase-signaling pathway. Hepatology 55, 1485–538 1494 (2012). 539

62. Donner, M. G. & Keppler, D. Up-regulation of basolateral multidrug resistance 540 protein 3 (Mrp3) in cholestatic rat liver. Hepatology 34, 351–359 (2001). 541

63. Hoehme, S. et al. Model Prediction and Validation of an Order Mechanism 542 Controlling the Spatiotemporal Phenotype of Early Hepatocellular Carcinoma. Bull 543 Math Biol 1–38 (2018). doi:10.1007/s11538-017-0375-1 544

64. Menshykau, D. et al. Image-based modeling of kidney branching morphogenesis 545 reveals GDNF-RET based Turing-type mechanism and pattern-modulating WNT11 546 feedback. Nature Communications 10, 239 (2019). 547

65. Uhl, E. W. & Warner, N. J. Mouse Models as Predictors of Human Responses: 548 Evolutionary Medicine. Curr Pathobiol Rep 3, 219–223 (2015). 549

66. Lau, J. K. C., Zhang, X. & Yu, J. Animal models of non-alcoholic fatty liver disease: 550 current perspectives and recent advances. J. Pathol. 241, 36–44 (2017). 551

67. Whitcomb, D. C. What is personalized medicine and what should it replace? Nat 552 Rev Gastroenterol Hepatol 9, 418–424 (2012). 553

68. Amalou, H. & Wood, B. J. Biopsy and personalized medicine. Nat Rev 554 Gastroenterol Hepatol 9, 683–author reply 683 (2012). 555

556


https://doi.org/10.1101/572073


557

Fig. 1. 3D reconstruction and quantitative analysis of human liver morphology. Human 558 liver sections obtained by biopsy (~100 um thick) were stained for bile canaliculi (CD13), 559 sinusoids (fibronectin), nucleus (DAPI), lipid droplets (BODIPY) and cell border (LDLR), 560 optically cleared with SeeDB and imaged at high resolution using multiphoton microscopy 561 (0.3 µm x 0.3 µm x 0.3 µm per voxel). For each sample, we reconstructed the central vein 562 (light blue), portal vein (orange), bile canaliculus (green), sinusoids (magenta), lipid droplets 563 (red), nuclei (random colours) and hepatocytes (random colours). a, Normal control. b, 564 Healthy obese. c, Steatosis. d, Early NASH. 565


https://doi.org/10.1101/572073


566

Fig. 2. Quantitative characterization of LD along the CV-PV axis. a, Representative IF 567 images of fixed human liver tissue sections stained with BODIPY. Shown is a maximum 568 projection of a 60 µm z-stack covering an entire CV-PV axis. b, Quantification of the 569 percentage of tissue volume occupied by the LD along the CV-PV axis and the overall values 570 (i.e. over the whole CV-PV axis). c, LD volume distribution. LD populations were defined 571 based on the LD volume distribution of the NC group (black line). Three populations of LD 572 were defined based on their volume: small (< 14 µm3), medium (14 – 400 µm3) and large (> 573 400 µm3), corresponding to changes in the volume distribution (i.e. drops in the number of 574 LD). Quantification of the percentage of tissue volume occupied by the LD along the CV-PV 575 for (d) small, (e) medium and (f) large LD. NC = 3 samples, HO = 3 samples, STEA = 3 576 samples, eNASH = 3 samples. Spatially-resolved quantification represented by mean ± SEM 577 per zone and overall quantifications by box-plots. *p-values < 0.05, **p-values < 0.01, ***p-578 values < 0.001. 579


https://doi.org/10.1101/572073


580

Figure 3. Cell based analysis of NAFLD. Quantification of the number of hepatocytes per 581 tissue volume unit (a) and cell volume (b) both, along the liver lobule and the overall average. 582 c, Cell volume distribution. For the population analysis, the hepatocytes form all the groups 583 were pulled together and the populations were defined based on their volume distribution 584 (c). By fitting the volume distribution with two normal distributions (c), the volume values 585 defining three population’s boundaries were identified: small (< 5800 µm3) (d), medium 586 (5800 – 11000 µm3) (e) and large (> 11000 µm3) (f). The fraction of cellular volume occupied 587 by the different populations is shown in d, e, and f. Percentage of the cell volume occupied 588 by lipid droplets: distribution (g) and statistics along the CV-PV axis and overall (h). NC = 3 589 samples, HO = 3 samples, STEA = 3 samples, eNASH = 3 samples. Spatially-resolved 590 quantification represented by mean ± SEM per zone and overall quantifications by box-plots. 591 *p-values < 0.05, **p-values < 0.01, ***p-values < 0.001. Representative cells reconstructed 592 in 3D and selected from zone 3 and 8. Apical, basal and lateral surface are shown in green, 593 magenta and grey respectively. Lipid droplets are shown in red. Scale bar, 10 µm. 594


https://doi.org/10.1101/572073


595

Figure 4. Structural and topological defects of bile canaliculi revealed by spatial 3D 596 analysis. a, Representative IF images of fixed human liver tissue sections stained with 597 CD13 after citric acid antigen retrieval. Shown is a maximum projection of a 60 µm z-stack 598 covering an entire CV-PV axis. b, Inset showing 3D representation of the bile canaliculus 599 highlighted in a. Quantification of the volume fraction of tissue occupied by bile canaliculi(c), 600 radius (d), number of junctions (e), total length per volume (f), fraction of connected network 601 (g) and connectivity density (h) of the BC network along the CV-PV axis and overall (See 602 Methods for details). NC = 5 samples, HO = 3 samples, STEA = 6 samples, eNASH = 5 603 samples. Spatially-resolved quantification represented by mean ± SEM per zone and overall 604 quantifications by box-plots. *p-values < 0.05, **p-values < 0.01, ***p-values < 0.001. 605


https://doi.org/10.1101/572073


606

Figure 5. Individual-based model prediction of bile pressure p and flow velocity 𝒗##⃗ 607 profiles based on measured bile canalicular geometries. a, Abstraction of liver lobule by 608 cylinder symmetry with radial coordinate r. The mechanistic model considers secretion of 609

osmolytes (green) and osmotic water influx (blue) in a porous medium with r-dependent 610 properties (see supplemental model description). Darcy’s law is assumed with a 611 proportionality constant K(r) depending on viscosity µ, tortuosity t, bile canalicular volume 612

fraction eBC, bile canalicular radius rBC. All geometric parameters have been measured per 613 patient. b-e, Model prediction for bile fluid pressure (solid line, left axis) and bile flow velocity 614 (dashed line, right axis) profiles for individual patients (colour) in for disease groups. NC = 5 615 samples, HO = 3 samples, STEA = 6 samples, eNASH = 5 samples. f, Scatter plot of 616 measured Gamma glutamyl transpeptidase (GGT) levels versus predicted pericentral (zone 617 0) bile fluid pressure from individual patients from all groups reveals a statistically significant 618 positive correlation. One-sided t-test. P-values and Pearson correlation coefficient are 619 indicated in the plots. 620


https://doi.org/10.1101/572073


3d spatially-resolved geometrical and functional models of ...66 recent advances in tissue staining,...

Documents